If you want a single number for English, here are two. About 3,000 high-frequency word families gets you to roughly 95% coverage of spoken English, which is the point where conversation stops being a guessing game. About 8,000 to 9,000 word families gets you to roughly 98% coverage of written English, which is the point where novels and newspapers become genuinely comfortable. Both figures come from Paul Nation's coverage research (Nation, 2006), and both are ranges rather than finish lines, because "fluent" means different things for talking, reading and working.
What fluency means in coverage terms#
Vocabulary research avoids the word "fluent" and measures coverage instead: the share of running words in a text that a reader already knows. Two thresholds come up again and again.
- 95% coverage is the threshold Laufer (1989) associated with adequate comprehension, and it is now generally treated as the floor at which a text works when some support is available, such as a glossary or a patient conversation partner. That is one unknown word in twenty.
- 98% coverage is where most learners read independently and for pleasure, without stopping (Hu and Nation, 2000). That is one unknown word in fifty.
Three percentage points sounds small. In vocabulary terms it is the difference between a few thousand word families and nearly ten thousand, which is why people can chat happily in English years before they can read a novel without effort.
What counts as one word#
Any English word count is meaningless until you know the unit being counted. The same page of text gives four very different numbers.
| Unit | What it means | English example |
|---|---|---|
| Token | Every running word, counted each time it appears | speak, speak, speaking is 3 tokens |
| Type | Each distinct written form, counted once | speak, speaking is 2 types |
| Lemma | A base word plus its inflections | speak covers speaks, spoke, spoken, speaking |
| Word family | A lemma plus its transparent derivations | speak also covers speaker, speech, unspoken |
English inflects far less than Spanish, German or Russian, so types and lemmas are closer together here than in most languages: a verb has a handful of forms, not dozens. The gap that matters in English is between lemmas and families: once you know nation, the forms national, nationally and nationality often come cheaply, which is the assumption behind family-based counts, though the saving is smaller for learners than the unit implies. When a study says "8,000 word families", it does not mean 8,000 separate memorisation events.
The coverage numbers for English#
Nation's 2006 study used the British National Corpus, a collection of around 100 million words of British English, and asked how many word families a learner needs for each coverage level. The anchor points below are from that work. The rows in between are approximations that follow the same curve, so read them as the shape of the problem rather than as measurements.
| Word families known | Written English (books, news) | Spoken English (conversation, TV) |
|---|---|---|
| 1,000 | around 80% | around 80-85% |
| 2,000 | around 90% | around 90% |
| 3,000 | low 90s | about 95% (Nation, 2006) |
| 4,000-5,000 | about 95% (Nation, 2006) | around 96-97% |
| 6,000-7,000 | around 97% | about 98% (Nation, 2006) |
| 8,000-9,000 | about 98% (Nation, 2006) | above 98% |
Two caveats come with those figures. Nation's counts assume that proper nouns and a handful of marginal items such as interjections are already known, which is a fair assumption for most readers. And speech is consistently easier than writing: conversation recycles a smaller core, while written English reaches for the rarer half of the dictionary.
Why the first 1,000 words do most of the work#
English word frequency follows the lopsided pattern formalised as Zipf's law (Zipf, 1949): a tiny set of words appears constantly and an enormous set appears almost never. In running English text, the single word the alone accounts for somewhere around five to seven percent of all words, and the hundred most frequent words together account for roughly half of everything written.
The practical effect is that early vocabulary is the best investment you will ever make in English. The top band is dominated by function words and general-purpose verbs: be, have, do, get, make, go, take, come, see, know. Michael West's General Service List (1953) captured this insight for English teaching with about 2,000 word families, and the idea still underpins modern lists built from the BNC and from COCA, the Corpus of Contemporary American English. Between them, those two corpora cover British and American English, and the high-frequency core is close enough across varieties that the top bands look much the same wherever you learn.
The same curve explains the bad news. Getting from 95% to 98% means learning thousands of words that each appear rarely, so the reps come slowly and you cannot force them. That stretch is the intermediate plateau in vocabulary form.
A huge dictionary and a small core#
English is famous for its size. The Oxford English Dictionary documents more than 600,000 words across a thousand years, including obsolete ones, partly because English absorbed vocabulary from Old Norse, Norman French and Latin and kept the overlaps. That history produced doublets with nearly the same meaning at different registers: begin and commence, freedom and liberty, help and assist.
That total matters much less to a learner than it looks. The rare half of the dictionary is rare for everyone. Estimates of what native speakers actually know vary with the counting unit: Goulden, Nation and Read (1990) estimated around 17,000 base word families for educated adult native speakers, while Brysbaert and colleagues (2016) estimated that an average twenty-year-old American knows roughly 42,000 lemmas. Those numbers are not in conflict so much as counting different things, which is the entire problem with vocabulary targets and the reason measuring your vocabulary size needs care.
What a learner needs is the shared core, and the core is small. Specialist vocabulary is a separate, smaller job: Coxhead's Academic Word List (2000) isolates 570 word families that recur widely across academic writing in the arts, commerce, law and science, sitting on top of the first 2,000 high-frequency families, which makes it a short list with a large payoff if you study or work in English.
What each CEFR level roughly needs#
The Common European Framework specifies no word counts at all, by design, since it has to work for every language. The ranges below come from exam-board profiling work and language-specific research, and sources disagree substantially, so treat them as orientation rather than targets. If the levels themselves are new to you, CEFR levels explained covers what each band claims.
| Level | Approximate vocabulary (estimates vary; sources differ on whether these are words or families) | What reading looks like |
|---|---|---|
| A1 | ~500-1,000 | Signs, menus, short predictable sentences |
| A2 | ~1,000-2,000 | Simple personal texts, low-level graded readers |
| B1 | ~2,000-3,000 | Simplified news, graded readers, factual texts with support |
| B2 | ~3,000-5,000 | Newspaper articles and professional material in a familiar field |
| C1 | ~5,000-8,000 | Literary fiction, dense journalism, academic writing |
| C2 | ~8,000-16,000 | Anything, including wordplay, dialect and unfamiliar technical fields |
For English specifically, Cambridge's English Vocabulary Profile, part of the English Profile programme, maps individual words and even individual meanings of a word to CEFR levels using learner corpus evidence. It is the closest thing English has to an official level-by-level word list, and it is worth a look if you want to know whether a given word is A2 or B2 rather than guessing. On time rather than words, Cambridge English publishes approximate guided-learning-hours guidance, running from a couple of hundred hours to reach A2 up to something over a thousand cumulative hours to approach C2. Those are order-of-magnitude figures that vary enormously with how close your first language is to English.
The things word counts quietly miss#
Three features of English make a raw family count optimistic.
Phrasal verbs. Put off, get over, look after and turn down are built from words you already know and mean something you cannot predict. Martinez and Schmitt (2012) assembled a list of roughly 500 frequent multiword expressions precisely because single-word lists leave them invisible, and many of them sit inside the most common English you will ever hear.
Polysemy. High-frequency English words carry many meanings. Run and get each carry hundreds of distinct senses in a full dictionary, so "knowing" them is not a single event. This is why meeting words repeatedly in different contexts beats meeting them once on a card.
Spelling and sound. English spelling maps onto pronunciation irregularly, so recognising a word on the page and recognising it in speech are partly separate skills. Learners who read widely often understand far more than they can follow in conversation, and the fix is listening volume rather than more words.
How to actually build them#
Two mechanisms have the strongest evidence behind them, and they work best together.
Comprehensible input. Krashen's input hypothesis (1985) holds that we acquire language by understanding messages slightly beyond our current level. For English this is easy to arrange, since there is more English material online than for any other language, but the difficulty step is the problem: a news site sits far above A2 vocabulary. Graded readers are the classic bridge, and the major series from Oxford and Penguin publish level charts running from a few hundred headwords up to a few thousand, so check the publisher's current chart and start a level below where your pride says you should. The underlying logic is covered in comprehensible input explained and in how to build vocabulary by reading.
Spaced repetition. Reviewing a word at expanding intervals fights the forgetting curve described by Ebbinghaus (1885) far more efficiently than massed review. The practical version is covered in spaced repetition for language learning.
LingoBlend is built around running both at once. You paste any text you actually want to read and choose what share of its words gets blended into English, from a light 5% up to full immersion, so the difficulty step is something you set rather than something the text imposes. Tap a blended word to see its meaning and grammar and save it, and saved words move into five practice games on a spaced-repetition schedule. For English specifically the Pro plan includes a few unabridged classics, among them Pride and Prejudice and The Great Gatsby, which are a reasonable place to go once simplified texts start feeling slow. The English learning guide is the starting point, and if you want to know where your own count currently stands, how many words do you know explains what a trustworthy counter can and cannot tell you.